Player is loading...
Digital detection of drug adverse events in rheumatology - a whole population analysis
Paul Hasler
Friday, Feb. 1, 2019 · 2:57 p.m. · 19m 33s
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
very much for the introduction and also thomas thank you very much for organising
00:00:04
this one girl caught wonderful conference uh it's beyond expectations thank you
00:00:09
so ah here we go uh with side effects of drugs
00:00:14
uh the bad news uh in a medicine and i like to lead you through uh with a short introduction
00:00:21
and then um lose a couple of words on data access address ability uh shows some clinical data
00:00:27
and uh how to extract salient features from the clinical data and uh give
00:00:33
a couple of examples and then the outlook where we go from here
00:00:37
so uh we know that i've comes or the core target of medical diagnostics and therapies
00:00:43
and it's always a question we've heard some uh uh questions about this today which data do we really rely on
00:00:51
and uh my big issue is how can we operational lies
00:00:54
our data to austin on to decide questions about outcomes
00:00:58
i'm gonna talk about data here i mean data from routine clinical cactus
00:01:04
so um there was a series of articles in nature uh not quite three years ago
00:01:09
uh about uh the drug industry and uh it's a a clear that we
00:01:15
show uh the drugs work before selling them and uh here it is
00:01:19
uh the spectrum deregulation and the office here cautioned against deregulation this here yeah
00:01:25
and even libertarian politicians in the u. k. like mac really
00:01:29
caution against the regulation uh of her drug testing
00:01:33
so um you want to drugs on the market um crowd you follow up on the drug is it really
00:01:41
effective a war uh is it really safe and we do have a issues uh that
00:01:46
we can solve with registries uh but i am a here is to uh
00:01:52
usually structured data or too structured data so that one can perform
00:01:56
post marketing studies that specifically to target organ diseases and also
00:02:02
off label use of drugs drugs so uh this is an area
00:02:06
which is very important and which we have very very few
00:02:09
data on uh as i said except for the registries which are the best a way that we can track this
00:02:15
now if we're in a department and we want to uh for for quality assurance uh tests
00:02:23
how effective are drugs in real life and have frequent are on desired
00:02:28
effects of drugs in routine here in my department and other departments
00:02:32
and also our potential serious truck engine interactions considered in routine prescription
00:02:38
uh for instance who considers a that patients navy on the staten when they prescribe call to see
00:02:45
this is a very important question and which that and you prescribe if
00:02:49
you're providing call to see what you need to change patterns
00:02:53
also which sequence and there wasn't of drugs is optimal for have
00:02:56
come we have huge databases of which you cannot address
00:03:00
and we do not know the answer to the question uh en route arthritis for instance if you start with one drug and
00:03:06
have a sequence of five next drugs uh will be do better uh with the if we change that sequence around
00:03:12
so this is uh the these are unanswered questions that could have a huge impact on the way we treat patients
00:03:20
so this is how conventional archive looks this was our condo archive
00:03:24
uh in our i have five years ago and uh we um
00:03:29
uh consider that this is a very good for routine here that looking for a bass
00:03:34
twenty eights in this uh uh in such a system is very very difficult
00:03:38
and uh this is essentially a data sick we take our notes we write reports and everything gets filed away
00:03:44
and uh filed away to be forgotten after ten years uh it gets shredded if
00:03:48
the patient doesn't come uh to the whole school again within that time
00:03:53
do electronic data systems uh provide relief from this dilemma
00:03:57
i mean we have here entry system where you can spend lots of time entering data you can't
00:04:02
even get a report had of it uh but you've done your work at the machine
00:04:06
and uh in the and uh if you can't address the data this is also a days
00:04:11
as think uh just like the paper data and job you mentioned that before work
00:04:15
uh that it's a terrifying that you have an electronic database and you can't even address the data that are in there
00:04:22
so um we set had uh in r. l. i'm following um these premises that
00:04:28
we wanted and unrestricted ability to migrate programs
00:04:32
and databases on to new platforms
00:04:35
uh this is because uh we had several systems from which we couldn't even extractor will be ports at
00:04:41
the hospital and my experience and also before i moved to our i was exactly the same thing
00:04:46
what are the problems was that uh these systems had propriety years in them
00:04:51
could not get through to address your data so no proprietary stored systems
00:04:56
and then of course would s. ability of data at blinding speed uh this is
00:05:01
one of the google a premises uh don't you don't should not have to
00:05:05
wait more than point three seconds for cancer uh if you query your system so
00:05:10
what is the solution that we had we scanned our whole our archive
00:05:14
uh we um use optical character recognition with pixel identical bit napping
00:05:20
uh of uh these scans uh to identify the letters the words
00:05:25
and uh we wanted to use this uh to have an automated recognition of
00:05:31
the case from each sheet of paper that went into the scanning system
00:05:35
so uh if stephanie brought souls uh she comes in for instance uh then the system
00:05:41
would recognise stephanie brought so this is the this is this case and a
00:05:45
ninety percent ninety seven percent correctness it would then um file this uh that the
00:05:50
sheet of paper or whatever it is uh into uh the a database
00:05:56
uh so this is a off the internet increments cans with automated case recognition
00:06:01
and the solution that was has the selected was uh uh no
00:06:06
s. q. l. data storage and and enterprise search engine
00:06:10
a search ability of the database with the apache loose installer uh when you book a
00:06:14
flight uh or you book your hotel uh or you or something i'm as on
00:06:19
uh these uh um up uh these operations are always using a some type of
00:06:25
search engine uh to um uh onto your queries and get your your best
00:06:31
a room at the ritz and caress uh if you chose choose to go to paris
00:06:35
for instance so what kind of data do we have had a roommate holiday department
00:06:39
uh we have almost thirteen thousand patient files that we've uh we have addressable
00:06:44
uh these are over two and a half million stand pages uh as
00:06:48
i said with an o. s. q. r. database installer search ability
00:06:51
and uh we can address the these uh uh the whole database with searches but
00:06:57
we can also search within individual cases so for instance if i'm looking for
00:07:01
prescription of a drug i can know back in the name of the drug
00:07:05
and prescription and then the drug will come out on the prescription sheet
00:07:11
so have questions looking lost here uh we had we looked at
00:07:15
office internet how show a couple of uh data on that
00:07:19
and uh this is uh the considered of potential interactions of coal
00:07:23
to see in and little felix titans in routine prescription
00:07:27
how many of our physicians uh to uh consider uh that they're a patients have a
00:07:33
potentially harmful pretty medication uh which will interact with
00:07:37
coaches scene then a method lexical rate toxicity
00:07:42
other toxicity symmetric sate liver toxicity which is more
00:07:45
frequent than probably from a a a toxicity
00:07:49
and uh something that we also have an interest in is outcome of john sell our try to so that we can
00:07:55
if the cat are cases of john sell arthritis and then a profile them clinically how does this work
00:08:01
uh we have this uh o. c. r. um sir system you can put
00:08:05
in your search word for instance that will chance effort to visit net
00:08:09
um you can put hands l. jensen room that and then you can see uh
00:08:13
we have eight uh eighty patients by now they're well over two hundred
00:08:17
uh that you can find with the system and then you can put in the l. chance right sept
00:08:21
uh uh a prescription and you'll find a prague uh fifty uh patients and these are uh
00:08:27
precise uh searches and we'll see that the term is i found on
00:08:32
or the combination of terms is found on a hundred and fifty five
00:08:35
scans so this is what we can uh do to identify patients
00:08:40
our cases and uh with that uh they can then be at the list of patients can be profiled
00:08:46
and individual uh characteristics of these patients uh
00:08:51
put into the table for statistical evaluation
00:08:56
and what we can do a is then that together this is uh with me you can miller
00:09:01
uh who did this analysis and in some tall and this is the now analysis of
00:09:05
a hundred and forty four patients uh and you can see why it's itself chance or
00:09:09
to visit neighbours stopped uh in if you can see that the most frequent reason
00:09:14
um you can see the the median day to a stop in due to even if they can see a is around two hundred then
00:09:21
you have gastrointestinal a side effects and then a side effects that become
00:09:26
a more more infrequent uh but with this you can also
00:09:30
find very very rare side effects uh within your whole database
00:09:35
so um then to call just seen so uh what what we do with call to sing we know the minorities
00:09:41
uh are increased control to see his code minister uh with little felix patterns
00:09:46
and uh so what we're going to do is we're going to uh set up a protocol
00:09:50
uh where we can uh search at all cases where coaches in isn't a is mentioned
00:09:55
uh that's um in our case in our uh over to six hundred and twenty cases
00:10:00
and in whom it was prescribed and also uh we will then look
00:10:04
for existing prescriptions of a taurus that understand the staff and
00:10:08
oh or think i'm new prescriptions which would be even worse
00:10:12
yeah and uh we can then see what's the staten
00:10:16
staten discontinued can call to sing was prescriber was an alternative staten prescribe
00:10:21
uh so this is a very simple question that after a routine practised uh that in the medication which is
00:10:27
a off label would not be possible to address a user the costs uh if you use other systems
00:10:34
so that to a message lexical re toxicity we all know about probably toxicity the frequency that
00:10:41
is given in the publications is point two from point three percent to over thirty percent
00:10:46
in cases that get method checks eight and here are the uh kramer criteria from nineteen ninety seven
00:10:53
uh that we can use it to identify these patients uh and
00:10:57
uh then you have cousins of major criteria on one
00:11:01
uh which is uh because the puddle pathological uh data or major criteria to and
00:11:06
free radio logic evidence supple read the special or how the older infiltrate
00:11:11
uh and then a block or in this just putting cultures negative for pathogenic organisms
00:11:17
that's so these are the three major criteria and then we have minor criteria
00:11:22
and then you can classify the patients as a definite or probable
00:11:27
um uh pull re toxicity due to the drought metrics eight
00:11:30
this is a bit superseded up here uh because at that time uh we
00:11:35
did not have data on broken i'll be all the lab arches
00:11:38
uh so uh we can use the this system as a screen for a database and
00:11:44
then we can then uh adjudicate cases and the pull disease specialist and then say
00:11:49
based on computerised demography uh off the longs and also the uh alveolar large results
00:11:56
uh what uh the definitive uh classification is these uh uh a
00:12:01
a list of search terms that we can use to
00:12:04
make certain that we do have all the cases uh with
00:12:07
metric say and potential pulley a side effects for evaluation
00:12:12
and if you look at the flow chart here we have almost two thousand cases of metrics it mentioned in the database
00:12:18
that we have a over a thousand six hundred that
00:12:21
metrics exposures and a seventh twenty seven cases
00:12:26
uh fulfil the creamer criteria exactly the same amount of uh patients that the six uh
00:12:32
in is that participated in the elaboration of the tray a trainer criteria had at
00:12:36
the time and these were then uh adjudicated by the pool were disease specialist
00:12:41
and in the end uh we have eleven cases with definite probably toxicity
00:12:46
three pays cases with probable up only toxicity uh
00:12:50
so fourteen all uh and that amounts to
00:12:53
i'm in a point eight percent uh all the cases uh that were uh exposed
00:12:59
so this is a possible uh application uh of a searchable
00:13:04
database uh in um uh like we've done in our
00:13:08
so where do we go from here i mean this is one single centre uh it's nice to have
00:13:12
a a nice number of cases um fourteen cases would pull re toxicity uh due to matter
00:13:18
checks eight and uh we uh have uh from the star which is about here
00:13:25
um and our our some call in and tell it so now uh put in the ground for a
00:13:33
system that we can uh look at these issues in several hospitals so
00:13:38
uh this is this create prima uh today there's gonna be a press
00:13:41
release from the controls but helpless along uh in the style
00:13:45
with your lucky uh as the lead a of this uh consortiums
00:13:50
and here uh uh this uh the uh the topic of this uh project uh off as p. h. and
00:13:56
uh is uh to create the infrastructure so that we can uh crossed
00:14:01
his situations uh for evaluations like i've told before and in
00:14:06
controls with ah ah so long as we have enterprise search engine address
00:14:10
ability of practically all the data in the hospital they have
00:14:14
three different clinical information systems they're a hundred and eighty thousand cases
00:14:18
in there over three and a half million patient reports
00:14:21
and six hundred million little retrieve data points and so what we're gonna do is we're gonna take our pulmonary disease toxicity
00:14:28
uh evaluation and go into a controls b. tell the style
00:14:32
and then a second phase will be able to go to the controls but also gallon which also has
00:14:37
a special t. with allah g. unit and also into another language area uh in bellied so now
00:14:43
and uh that part of this uh will be our uh twelve thousand
00:14:47
eight hundred uh and drawing room apology cases in our our
00:14:50
and our work tree is considerably larger than the one in the style
00:14:54
uh with three mill three and a half million a data point
00:14:58
sorry three and a half billion data points of almost three quarter of a million cases and for instance you
00:15:03
can search the whole database uh with criteria for love or three parameters like trans iman is is
00:15:09
and you can also look at this real time so if we have a link of a patient with metrics eight
00:15:14
and we see that the trans am races in the lab and the lab rise uh we can have a flags
00:15:19
so that's um uh the future uh but first we're going
00:15:22
to look at our uh metric say toxicity cases
00:15:26
uh within this database in the style and then go uh to some call and and to uh bellied so now
00:15:33
so to summarise search engine access and analysis of real world medical data is feasible uh one kind
00:15:40
identify and evaluate cases for clinical and study purposes
00:15:44
uh this is a very important uh also for very rare diseases or
00:15:49
very rare um the side effects of for instance not objects eight
00:15:54
causes epilepsy uh in that extremely low amount of patients but we do
00:15:58
have a case that has an epilepsy due to metrics aid and
00:16:03
um we can uh use uh this a database for intelligent assistance
00:16:09
uh to uh find these uh uh side effects and also evaluate the frequency of that they have
00:16:15
uh and as i said uh the expansion over different data sets and other hospitals is imminent
00:16:20
so thank you for your tension ha
00:16:26
very nice kirk there are there any question for price so hard for
00:16:32
it's so time
00:16:36
so what i get an a warning um
00:16:40
when the system detects antiques toxicity somewhere
00:16:46
if they in your department or elsewhere up this is not definitely not the the the uh
00:16:52
issue right at the moment but uh for instance at least half where the
00:16:55
love or tree data or operational lies and you have real time
00:16:59
uh accessibility uh if you set a troubling over two hundred we'll get on to the
00:17:04
uh directly onto the smart phone of the cardiologists uh who was on call
00:17:08
and uh this will even happen from the ambulance if it's from the point of care testing a gadget uh
00:17:13
from the lady infusion they take the blood uh and uh twelve minutes later you have this problem
00:17:19
so that's a uh that's possible with the system they haven't these tile
00:17:24
unfortunately we started out and are on the style is far far ahead of what we are and i think that has
00:17:30
several uh reasons one reason is that our hospital is also larger listening more difficult to get anything but through
00:17:38
i have one question for for people who are not doctorate here
00:17:45
has to not have to know that they're directory t. there an
00:17:48
adverse effects usually have to be reported to systematic so
00:17:53
you decide to to you know buy low but they're far
00:17:56
fishermen take serve thirty minutes more or even more
00:18:00
to report to fill in paper form aries and do you think your system
00:18:04
and process could be used to automatically reports to cease maybe care
00:18:10
uh that's a very good question that one thing i didn't mention here is that we have a semi automated
00:18:16
in port of call medications and a co morbidity is
00:18:20
uh into this table uh and uh once the
00:18:24
lab data are accessible in our our uh we have discrete data points that are time stamped
00:18:30
so we can construct a sequence of events uh based on the lab data
00:18:36
even if there are gaps in the lab data and this is definitely something that can be uh that can be uh
00:18:41
automated uh so you have a trend salmon is is you have the whole the lab data set you can actually
00:18:47
transfers just medic if they want to have all that data uh and they can
00:18:51
evaluate and so this is a this is all it's also possible to completely
00:18:56
d. personalise the data to so that will be in you can take all
00:18:59
identify as operations and also treating physicians uh practitioners out of the system
00:19:07
thank you so we move around my room place and to welcome a doctor you i'm sure oh uh
00:19:13
working out there the repetition inner can your your mystical speak now and she will talk about the
00:19:20
impact of her a patient apps are on quality of
00:19:23
care and you know that quality of koreans are
Conference Program
Demystification the digital health world
Thomas Hügle
Feb. 1, 2019 · 9:41 a.m.
6530 views
Digital Needs, fears and possible solutions in rheumatology - From a patients' perspective
Janos G. Kozma
Feb. 1, 2019 · 10:28 a.m.
1296 views
Digital RA - prediction and therapy by artificial intelligence?
Jacob Van Laar
Feb. 1, 2019 · 10:45 a.m.
1405 views
Digital lessons learned from musculoskeletal radiology
Patrick Omoumi
Feb. 1, 2019 · 11:37 a.m.
1896 views
Big data: possibilities and threats for complex disease genetics
Zoltan Kutalik
Feb. 1, 2019 · 12:04 p.m.
1349 views
Developine fine-grained actigraphy's for rheumathoid arthiritis patients from a single accelerometer using machine learning, sensors
Javier Andreau Perez
Feb. 1, 2019 · 1:51 p.m.
1180 views
Deep learning meets rheumatology: new methods to predict disease progression in arthritis
Maria Hügle
Feb. 1, 2019 · 2:38 p.m.
1399 views
Digital detection of drug adverse events in rheumatology - a whole population analysis
Paul Hasler
Feb. 1, 2019 · 2:57 p.m.
1146 views
Effect of patient app use on quality of care in rheumatic disease
Yomei Shaw
Feb. 1, 2019 · 3:16 p.m.
1194 views
Automated image analysis in RA
Richard Ljuhar
Feb. 1, 2019 · 4:04 p.m.
1184 views
ELECTOR - eHealth in rheumatology. Patients with rheumatoid arthritis acquire sustainable skills for home monitoring
Marie Skougaard
Feb. 1, 2019 · 4:27 p.m.
1141 views
Return of experience: what are the digital tools available to train HCPs
Xavier Abadie
Feb. 1, 2019 · 4:50 p.m.
1105 views
Differential diagnosis assessment in ambulatory care with the automated medical history-taking device "Diaana", supported by a pilot randomized study
Adrien Schwitzguebel
Feb. 1, 2019 · 5:31 p.m.
1255 views
Recommended talks
How nanofluidics bring diagnostics closer to the patient
Fabien Rebeaud, Abionic
Oct. 26, 2017 · 4:17 p.m.
224 views